Empir Econ (2011) 40:755–777

DOI 10.1007/s00181-010-0366-7

Estimating the natural rates in a simple New Keynesian

framework

Hilde C. Bjørnland · Kai Leitemo · Junior Maih

Received: 12 January 2008 / Accepted: 20 January 2010 / Published online: 8 April 2010
© Springer-Verlag 2010

Abstract The time-varying natural rate of interest and output and the implied
medium-term inflation target for the US economy are estimated over the period 1983–
2005. The estimation is conducted within the New Keynesian framework using Bayes-
ian and Kalman-filter estimation techniques. With the model-consistent estimate of
the output gap, we get a small weight on the backward-looking component of the
New Keynesian Phillips curve—similar to what is obtained in studies which use labor
share of income as a driver for inflation (e.g., Galí, Eur Econ Rev 45(7):1237–1270,
2001; Eur Econ Rev 47(4):759–760, 2003). The turning points of the business cycle
are nevertheless broadly consistent with those of CBO/NBER. We find considerable
variation in the natural rate of interest while the inflation target has been close to 2%
over the last decade.

Keywords Natural rate of interest · Natural rate of output · New Keynesian model ·
Inflation target

JEL Classification C51 · E32 · E37 · E52

H. C. Bjørnland (B) · K. Leitemo

Department of Economics, Norwegian School of Management (BI), Nydalsveien 37, 0442 Oslo,
Norway
e-mail: hilde.c.bjornland@bi.no
K. Leitemo
e-mail: kai.leitemo@bi.no

H. C. Bjørnland · J. Maih
Norges Bank, 0107 Oslo, Norway
e-mail: junior.maih@norges-bank.no

123
756 H. C. Bjørnland et al.

1 Introduction

The New Keynesian theory, as developed by Goodfriend and King (1997), Rotemberg
and Woodford (1997), McCallum and Nelson (1999), and others, and with policy
implications extensively explored in Clarida et al. (1999) and Woodford (2003), has
become the leading framework for the analysis of monetary policy. This theory honors
the proposition that monetary policy affects only nominal variables in the long run
and that the steady-state inflation rate can be governed by monetary policy. More-
over, it assumes that the central bank implements its policy through the setting of the
short-term interest rate. Monetary policy influences decisions about real magnitudes
due to prices not being fully free to adjust to shocks (price rigidities). The overriding
objective of monetary policy is to alleviate the effects of these rigidities while keeping
inflation expectations close to a target rate of inflation.
An important point of reference for the policymaker is how the economy would
have developed had prices been without rigidities and instead fully flexible. We refer
to the rate of interest and the level of output in such an equilibrium as the natural rates
of interest rates and the natural level of output (see Woodford 2003). Consistent with
this view, the strategy of monetary policy is often formulated in terms of deviations
from these natural rates, that is, in terms of the interest rate gap and the output gap,
respectively. The well-known Taylor rule (Taylor 1993) provides an illustration. Under
the Taylor rule, the central bank raises the interest rate relative to the natural rate of
interest if either inflation deviates from the inflation target and/or output deviates from
the natural level of output. For these reasons, the natural rates are important indicators
for the setting of the policy instrument and the characterization of a neutral monetary
policy stance.
The main objective of this article is to present a simple framework in which to
derive the natural rates within a New Keynesian model setting. The model is small,
yet incorporates the main ingredients of the New Keynesian framework, making it a
useful device of analyzing how changes in the natural rates affect the economy and
monetary policy. Despite the simple nature of the model, we derive plausible time-
varying estimates of the natural rates and the corresponding interest rate and output
gaps using Bayesian estimation and Kalman filtering techniques on the US data. Pre-
vious studies on the topic include the seminal article by Laubach and Williams (2003)
who use the Kalman filter to estimate the (unobserved) natural rate of interest and
the output gap. The model is a standard growth model, implying that the natural real
interest rate varies over time in response to shifts in preferences and trend growth
rate of output.1 Their models, however, specify the natural rates within a reduced
form system devoid of forward looking elements.2 In this regard, our article is more
related to the recent unobserved components study of Basistha and Nelson (2007)
who acknowledge that inflation may be dependent on expected future inflation. They
derive the output gap assuming that inflation depends on (survey measures of) expected
future inflation as well as past inflation rates and the output gap. We extend on their

1 The idea builds on the articles by Watson et al. (1997) and Gordon (1998), among others, which estimate
the natural rate of unemployment (NAIRU) using the Kalman filter.
2 See also Garnier and Wilhelmsen (2005) for an application to the Euro Area.

123
Natural rates in a simple New Keynesian framework 757

contributions by deriving the estimates in a way that is consistent with New Keynesian
theory restrictions and furthermore allowing inflation expectations to be rational and
theory consistent. Our approach is nevertheless spare relative to a full DSGE approach,
in that we neither impose technology restrictions nor model the market for produc-
tion factors. The supply side of the economy is governed by exogenous processes.
This approach allows us to have a relatively simple model while allegedly being less
sensitive to possible controversial assumptions required to model, e.g., marginal costs
explicitly.
Another facet of the contribution of this article is the allowance of the possibility of
a time-varying inflation target. The US inflation history is difficult to reconcile with
a constant inflation target. In this regard, our approach is similar to that of Ireland
(2007). While we, however, assume that the inflation target reflects a preference of the
monetary policymaker and is unrelated to the state of the economy, Ireland assumes
that the inflation target is dependent on some of the shocks to private sector behavior.
Our conclusions regarding the evolution of the inflation target are nevertheless similar.
A third novelty of our approach is that it does not require detrending of the data prior
to analysis (using for instance the HP-filter) or making output stationary by deflating
by a trending variable (for instance, by assuming that total factor productivity follows
a trend stationary process), as has been common practice in many recent DSGE analy-
ses, including Edge et al. (2007), Juillard et al. (2005), Andrés et al. (2005), and Smets
and Wouters (2003, 2007) who also estimate the natural rates.3
An important empirical finding in this article is that inflation is primarily a for-
ward-looking process. By allowing inflation to have both forward-looking and back-
ward-looking components, using a hybrid New Keynesian Phillips curve, data prefer
a forward-looking specification. Although this is the common conclusion in studies
which use labor’s share of income as a proxy for marginal costs (see Galí et al. 2001,
2003), it is not common finding when the output gap is the driving process. Interest-
ingly, after accounting for the time-varying inflation target and natural rate of interest,
a model-consistent estimate of the output gap gives rise to a Phillips curve specification
similar to that of labor’s share of income. We interpret this in favor of using the output
gap as a valid representation of the inflation driver. This suggests that the approach of
studying monetary policy within the a simple model framework with inflation, output
gap, and the interest rate, as advocated in Woodford (2003), is also empirically useful.
The remainder of this article is organized as follows. In Sect. 2, we present the New
Keynesian framework. Section 3 presents the estimation framework and results. In
Sect. 4, we provide some concluding remarks.

2 A simple New Keynesian framework

The New Keynesian framework assumes that firms operate in monopolistic competi-
tive markets and production is constrained by aggregate demand. Prices are assumed

3 A recent exception is Juillard et al. (2006). They allow for a more general stochastic process where there
could be both temporary changes in the growth rate of total factor productivity as well as autocorrelated
deviations from steady state.

123
758 H. C. Bjørnland et al.

to be sticky and consequently do not move instantaneously to movements in mar-

ginal costs. Owing to the price stickiness, the central bank affects aggregate demand
through its influence on real interest rates. By lowering real interest rates, the cen-
tral bank induces higher aggregate demand, marginal costs, and prices than would
otherwise materialize. As noted above, the natural rate of interest rate can be regarded
as the neutral stance of monetary policy—the real interest rate that produces zero
output gap and stable inflation.
In estimating the natural rates, we build on the economic structure provided by the
New Keynesian framework. The basic model is extended with external habit formation
in consumption (Fuhrer 2000) and a hybrid New Keynesian Phillips curve that allows
for both forward-looking and backward-looking elements. This set up is rationalized
by the Calvo (1983) framework with some of the firms setting prices in accordance
with an indexation scheme (Christiano et al. 2005) or in accordance with some rule-
of-thumb (Galí and Gertler 1999). Our approach remains, nevertheless, conservative
regarding the extent of the economic structure regarding production technology and
the structure of the labor market imposed in estimation. This reduces the approach’s
rigor at the gain of not being tied up to a particular description of production tech-
nology which may bias the result if incorrect. Specifically, we allow the natural rate
of output to follow exogenous autoregressive (AR) processes and in this regard, the
article draws on the literature on structural time-series estimation, see e.g., Harvey
(1989).

2.1 Aggregate demand

We assume that the economy consists of a representative household that lives forever
and maximizes expected utility given by

∞  i   (1−σ ) 
 1 1 Ct+i Vt+i
U = Et ,
1+δ (1 − σ ) Ht+i
i=0

subject to the intertemporal budget constraint given by

 
Mt Bt Wt Mt−1 Bt−1 Tt
Ct + + = Nt + + It−1 − + t .
Pt Pt Pt Pt Pt Pt

δ is the discount rate, σ is the intertemporal elasticity of substitution and C is an CES

index of consumption goods. V is a consumption preference shock. The consumer
is also assumed to have preferences over money and leisure. The decision processes
associated with labor supply decisions are not explicitly modeled and implicitly left
exogenous in the model. The reason for doing this is partly simplicity, and partly a
reflection of our view that the approaches currently available for modeling the labor
market decisions are too simplistic. Hence, imposing restrictions from these theories
are likely to be biasing our results. The cost of keeping the production technology
“exogenous,” however, is that we cannot distinguish between particular shocks on the
supply side, e.g., productivity versus mark-up shocks.

123
Natural rates in a simple New Keynesian framework 759

The consumer can either hold money (M) or bonds (B) as a store of wealth. Money
yields utility (not modeled) whereas bonds yield a gross risk-free return of It in every
period. Consumption preferences are subject to a shock Vt ≡ (1 − vt ) where

vt = ρv vt−1 + ṽt (1)

where ρv is degree of persistence in the shock and ṽt is a white-noise shock. Ht

represents external habit persistence. We introduce habit persistence of order 2. The
reason for this is that it allows for a higher-order lag structure of the resulting first-order
condition. The habit persistence is specified as follows:

γ γ
Ht = Ct−1
1 2
Ct−2 ,

where γ1 and γ2 are habit parameters. This more general setup allows agents to form
habits with respect to the changes in as well as the level of consumption.
The first-order condition for the solution to the problem implies the consumption
Euler equation

 1−σ    1−σ
Ct Vt 1 1 Ct+1 Vt+1 1 Pt
γ1 γ2 = It E t γ γ2 . (2)
Ct−1 Ct−2 Ct 1+δ Ct 1 Ct−1 Ct+1 Pt+1

Taking the logarithm of the Euler equation and using the resource constraint, we have

σ (γ1 − γ2 ) (σ − 1)
yt = E t yt+1 + yt−1
A A
γ2 (σ − 1) 1 (σ − 1)
+ yt−2 − (i t − E t πt+1 − δ) + (vt − E t vt+1 ) , (3)
A A A

where A ≡ σ + γ1 (σ − 1) and πt is quarterly inflation at an annual rate. A small

letter denotes the log of the corresponding capital letter variable.4
Note that due to dynamic homogeneity, we can write the aggregate demand schedule
(3) as

σ γ2
yt = E t yt+1 − yt−1
γ1 (σ − 1) γ1
1 1
− (i t − E t πt+1 − ρ) + (vt − E t vt+1 ) . (4)
γ1 (σ − 1) γ1

4 Note that we have for simplicity ignored Jensen’s inequality and used first-order Taylor approximations,
implying ln E(1 + x) = E ln(1 + x) = E x.

123
760 H. C. Bjørnland et al.

2.2 Aggregate supply

Aggregate supply is represented by the hybrid Phillips curve as


4
πt = μE t πt+1 + (1 − μ) α j πt− j + κ xt + εt , (5)
j=1

where (1 − μ) is the weight on the backward-looking component, εt is a cost push

shock and xt ≡ yt − ytn is the output gap, defined as the deviation of output from the
natural rate of output. As in Rudebusch (2002a,b), we allow for a lag structure on past
inflation to match the dynamics of inflation at the quarterly frequency. Furthermore,
we impose dynamic homogeneity, i.e., that α4 = 1 − α1 − α2 − α3 .5
As noted above, we do not endogenize the input of production factors and specify
technology, but instead assume that the natural rate of output is given exogenously by
the process

ytn = v + ωt (6)

where ν is the unconditional expected growth rate of output and ωt is an AR(1) shock
to the growth rate (natural rate shock)6

ωt = φωt−1 + t . (7)

The output gap then follows the process

xt = xt−1 + yt − ytn . (8)

2.3 Monetary policy

The monetary authority is setting the interest rate in accordance with a dynamic Taylor
rule as

i t = ψi t−1 + (1 − ψ) i tn + θπ π̄t − πtT + θx xt + u t , (9)

where ψ measures the smoothing in the interest rate setting. i tn is the nominal natural
interest rate (defined below) and

5 Although we do not provide any microfoundations for these lags, we postulate that these lags will follow
from the rules-of-thumb framework of pricing of Galí and Gertler (1999) given that rule-of-thumb allows
for longer lags.
6 The shock  is best thought of as representing variations in productivity and preferences that influence
the marginal rate of substitution between consumption and leisure. Neither sources is modeled explicitly
here.

123
Natural rates in a simple New Keynesian framework 761

1
3
π̄t ≡ πt− j
4
j=0

is the four-quarter inflation at an annual rate. We assume the FED has an implicit
intermediate-run target for inflation that can deviate from the long-run (steady state)
inflation target. This could play the role of smoothing inflation dynamics and bring
inflation slower back to target once above it, avoiding large output changes. It evolves
according to

πtT = (1 − ρπ ) π ∗ + ρπ πt−1
T
+ ξt , (10)

where π ∗ is the steady-state inflation rate (or long-run inflation target), and ξt is an
AR(1) shock to the inflation target, in accordance with

ξt = ρ ξt−1 + t . (11)

2.4 The natural rate of interest

The process for the natural nominal rate of interest can be found by replacing output
and the interest rate in Eq. 3 with the natural rates and then solving for the interest
rate, i.e.,

σ (γ1 − γ2 ) (σ − 1) n
ytn = n
E t yt+1 + yt−1
A A
γ2 (σ − 1) n 1 n  (σ − 1)
+ yt−2 − i t − E t πt+1 − δ + (vt − E t vt+1 ) , (12)
A A A

or

σ γ2 n
ytn = E t yt+1
n
− yt−1
γ1 (σ − 1) γ1
1 n  1
− i t − E t πt+1 − δ + (vt − E t vt+1 ) . (13)
γ1 (σ − 1) γ1

and isolating for the natural interest rate

i tn = δ + E t πt+1 + σ E t yt+1
n
− γ1 (σ − 1) ytn − γ2 (σ − 1)yt−1
n

+ (σ − 1) (vt − E t vt+1 ) . (14)

The natural real interest rate is then found from the Fisher equation as

rtn ≡ i tn − E t πt+1 . (15)

123
762 H. C. Bjørnland et al.

The output gap process can be expressed as a function of the natural interest rate
by subtracting Eq. 12 from Eq. 3 which gives

σ (γ1 − γ2 ) (σ − 1)
xt = E t xt+1 + xt−1
A A
γ2 (σ − 1) 1  
+ xt−2 − i t − i tn (16)
A A

where the natural rate of interest is given in Eq. 14 above.

3 Estimation

We estimate the parameters of the model comprising of Eqs. 1, 4, 5, 6, 7, 8, 9, 10,

and 11 using Bayesian methods and the Kalman filter. The focus of the analysis will
be on the estimation of the natural real rate of interest and the output gap. The use
of Bayesian methods to estimate DSGE models has increased over recent years, in a
variety of contexts, see An and Schorfheide (2006) for a recent evaluation. The focus is
on methods that are built around a likelihood function, typically derived from a DSGE
model (see, e.g., Adolfson et al. 2007). With sensible priors, Bayesian techniques offer
a major advantage over other system estimators such as maximum likelihood, which in
small samples can often allow key parameters to wander off in nonsensical directions.

3.1 Data

We estimate the model laid out in the previous section using the US quarterly time
series for three variables: real output, inflation, and interest rates. The sample period
is 1983q1–2005q4. The period covers the last part of the Volcker period and the major
part of the Greenspan period. The choice of periods follows from the assumption that
these two Chairmen shares approximately the same dislike for inflation. The monetary
policy regime is, therefore, roughly constant over the sample period. We use the quar-
terly average daily readings of the US 3-month deposit rates as the relevant nominal
interest rate. For real output and inflation, we use real GDP and the CPI, all items,
for the entire USA. GDP and CPI are seasonally adjusted by their original source
(OECD). We treat inflation, output growth, and the nominal interest rate as stationary,
and express them in deviations from their sample mean. Note that all the changes are
measured at an annual rate.

3.2 Parameter estimation

As is well known from Bayes’s rule, the posterior distribution of the parameters is pro-
portional to the product of the prior distribution of the parameters and the likelihood
function of the data. This prior distribution describes the available information prior
to observing the data used in the estimation. The observed data are then used to update
the prior, via Bayes theorem, to the posterior distribution of the model’s parameters.

123
Natural rates in a simple New Keynesian framework 763

Table 1 Estimation results for the US economy

Coefficients Prior mean Prior SD Distr. Support Post. mean 5% 90%

Phillips curve
μ 0.50 0.20 Beta [0, 1] 0.626 0.314 0.908
α1 0.25 0.10 Norm None 0.353 0.200 0.506
α2 0.25 0.10 Norm None 0.240 0.100 0.366
α3 0.25 0.10 Norm None 0.227 0.094 0.369
α4 0.25 n/a n/a n/a 0.180 n/a n/a
κ 0.20 0.15 Gamm [0, ∞] 0.089 0.005 0.163
IS curve
δ 0.04 0.02 Gamm [0, ∞] 0.016 0.006 0.027
σ 2.00 0.50 Beta [1.05, 5] 2.047 1.625 2.448
γ1 0.50 0.20 Beta [0, 1] 0.537 0.332 0.727
γ2 0.40 0.20 Beta [0, 1] 0.599 0.396 0.870
ρv 0.85 0.10 Beta [0, 1] 0.945 0.916 0.980
Natural rate process
φ 0.850 0.10 Beta [0, 1] 0.788 0.678 0.909
υ 0.030 0.005 Gamm [0, ∞] 0.029 0.024 0.035
Monetary policy
ρ pi 0.800 0.10 Beta [0, 1] 0.853 0.751 0.950
ρχ 0.800 0.10 Beta [0, 1] 0.795 0.662 0.939
θπ 0.500 0.10 Beta [0.1, 1.5] 0.578 0.420 0.720
θx 0.500 0.10 Beta [0.1, 1.5] 0.449 0.284 0.570
ψ 0.700 0.10 Beta [0, 1] 0.828 0.793 0.872
Standard deviations of shocks
σ 0.002 Inf Invg [0, ∞] 0.0024 0.0008 0.0045
σε 0.001 Inf Invg [0, ∞] 0.0110 0.0091 0.0127
σu 0.001 Inf Invg [0, ∞] 0.0024 0.0020 0.0028
σṽt 0.001 Inf Invg [0, ∞] 0.1983 0.1181 0.3045
σ 0.001 Inf Invg [0, ∞] 0.0096 0.0074 0.0119

In order to implement the Bayesian estimation method, we need to be able to

evaluate numerically the prior and the likelihood function. Then, we use the Metrop-
olis-Hastings algorithm to obtain random draws from the posterior distribution, from
which we obtain the relevant moments of the posterior distribution of the parameters.
More specifically, the model is estimated in two steps in Dynare-Matlab. In the first
step, we compute the posterior mode using ‘csminwel,’ an optimization routine devel-
oped by Christopher Sims. We use the first 3 years of the full sample 1983q1–2005q4
to obtain a prior on the unobserved state, and use the subsample 1986q1–2005q4 for
inference. In order to calculate the likelihood function of the observed variables, we
apply the Kalman filter. In the second step, we use the mode as a starting point to
compute the posterior distribution of the parameters and the marginal likelihood by
simulations of the Metropolis-Hasting (MH) algorithm (see for details Schorfheide

123
764 H. C. Bjørnland et al.

2000). The debugging features of Dynare are used to determine whether the optimiza-
tion routines have found the optimum and whether enough draws have been executed
for the posterior distributions to be accurate. Having estimated the parameters, they
can then be used to construct the natural rates of interest rates and output.

3.3 Prior and posterior distributions

The Bayesian estimation technique allows us to use prior information from previous
micro- and macro-based studies in a formal way. Table 1 summarizes the assumptions
for the prior distribution of the estimated parameters and structural shocks. In the
first three columns, the list of structural coefficients with their associated prior mean,
standard deviations and distribution are shown. On the basis of standard conventions,
we use Beta distributions for parameters that fall between zero and one, (inverted)
gamma (invg) distributions for parameters that need to be constrained to be greater
than zero, and normal (norm) distributions in other cases. For some of the parameters,
the distribution is constrained further, as indicated in column four (‘support’).
The next three columns indicate the posterior mean and the associated 90% uncer-
tainty interval. Starting with the Phillips curve, we provided a prior for μ = 0.50 which
puts equal weight on the forward-looking and backward-looking components with a
large standard deviation providing a rather diffuse prior. This choice is rationalized
by the fact that the literature has suggested estimates in the whole zero-unity interval.
We wanted data to determine this coefficient without pushing it in either direction. In
the estimation, α1 , α2 , α3 , and α4 were restricted to sum to one (with α4 determined
by this identity). However, since we do not have a strong prior on their magnitudes,
we give them the same weight with the standard deviation set to 0.1. κ was estimated
at 0.089 which is not far from the estimate of 0.13 obtained by Rudebusch (2002a,b)
who used CBO estimate of the output gap.
We find that the Phillips curve is primarily forward looking. It has nevertheless a
non-negligible weight on the backward-looking component with (1 − μ) just below
0.4. This is consistent with the estimates of the New Keynesian Phillips curve found
when using labor’s share of income as the proxy for marginal costs7 as opposed to
using detrended output. We believe that this result is due to allowing for simulta-
neous estimation of the Phillips curve parameter and the natural rates, which are both
closely connected. The use of detrended output in some studies disregards this impor-
tant simultaneity. Our results are consistent with the estimation results in Galí et al.
(2003, 2005) using a full information, system estimation. We find this result inter-
esting because it suggests that the output gap may be a valid representation of the
inflation driving process. Hence, modeling the measures of marginal costs may not
be essential to capture a broad representation of the monetary policy transmission
mechanism. The results support that monetary policy can be studied within a simple
two-equation model framework which explains the development of inflation and the
output gap conditional on the policy instrument (as suggested by Clarida et al. 1999
and Woodford 2003).

7 See, e.g., Galí et al. (2001, 2003, 2005) and Sbordone (2002, 2005).

123
Natural rates in a simple New Keynesian framework 765

Table 2 Model fit: standard deviations

Standard deviations Interest rate Inflation Output growth

Theoretical moments 0.0260 0.0144 0.0219

Actual sample moments 0.0269 0.0209 0.0206

Regarding the expectational IS curve, we find that our prior on the intertemporal
elasticity of substitution σ = 2 is well within the range of the estimates in the literature.
The posterior has increased somewhat from the prior, although not significantly so
(posterior mean equals 2.05). Moreover, the preference shocks display a high degree
of persistence, with a coefficient of ρv = 0.95. In addition, the habit parameters γ1
and γ2 are restricted to lie between zero and one, with the prior for γ1 being the largest,
assuming more habit from the immediate past. However, we choose a large standard
deviation that provides us with a fairly diffuse prior. The second-order habit persis-
tence is well accounted for in data, as both γ1 and γ2 turn out to be above the priors.
Finally, the prior for the annual discount rate δ is set to 0.04, reflecting a quarterly
discount factor of 0.99. Rather surprisingly, we find that data push the annual discount
rate from the prior of 4 to 1.6%.
The prior for the equilibrium natural output growth rate is set equal to the (annual)
growth rate in the model (3%), with the posterior mean estimated to υ = 0.029. As our
data set is small, it is unlikely that we would get any other value than the equilibrium
value suggested by the data. As an alternative, we could, therefore, have calibrated
this value at 0.029.
The data seem to support a dynamic Taylor rule specification of monetary policy
reasonably well. The monetary policy shock (σu ) has standard deviation of 0.024.
Moreover, the weights on inflation and output gap are deviating only marginally from
the priors and what Taylor (1993) suggested as likely coefficients (0.5). There is a pro-
nounced gradual adjustment of the interest rate with ψ = 0.83. Finally, we calibrate
the steady-state inflation rate π ∗ to be equal to steady state inflation. The results seems
to indicate fairly persistent movements in the medium-run inflation target (ρπ = 0.85),
with also rather persistent shocks to this process (ρχ = 0.80). The latter suggest that
movements in the medium-run inflation target is done gradually over time.
Finally, we note that the fit of the model seems to be reasonably good in terms of
matching moments. In the first row of Table 2, we show the sample moments (standard
deviations) from the smoothed posterior predictive distribution of the three observable
variables: interest rates, inflation, and output growth. The second row show the same
sample moments, but calculated from the actual US data. The estimated model seems
to fit reasonably well and is able to explain the larger part of the salient features of the
data as the actual sample moments does not lie too far from the posterior predictive
distribution.

3.4 Error variance decomposition and impulse responses

Table 3 shows the decomposition of the unconditional variance. Some interesting

observations can be made from the table. We first note that the main drivers of infla-

123
766 H. C. Bjørnland et al.

Table 3 Error variance decomposition

Variables and shocks Inf.-tar. () Cost-push (ε) Mon. pol. (u) Preference (ṽ) Nat. rate ()

rn 0.00 0.00 0.00 91.52 8.48

x 25.14 21.75 6.51 44.04 2.56
π 31.58 48.25 1.46 17.56 1.14
i 10.79 3.04 0.85 83.91 1.41
in 10.81 2.89 0.44 78.08 7.78
πT 100.00 0.00 0.00 0.00 0.00

tion variations are the cost-push and inflation-target shocks. These shocks account
for about 80% of the variation in inflation. If the central bank adheres to an infla-
tion-targeting loss specification with the loss function having inflation and output gap
variations as the two arguments (see Svensson 1997; Clarida et al. 1999), efficiency
in policymaking requires that inflation should be driven only by cost-push and infla-
tion-target shocks. The ratio is high and can be taken as an indication of efficiency in
policymaking. However, by the same logic, the central bank should fully neutralize
the impact of preference shocks on both the output gap and inflation. This does not
seem to be the case. Although the Taylor rule has allowed strong responses to the
preference shocks as they can explain more than 80% of the variation in the interest
rate, preference shocks have still influenced inflation and, in particular, the output gap
to a large extent. Hence, the estimated Taylor rule does less well in insulating the
economy from this type of shock.
The natural real interest rate is driven mainly by preference shocks that make
demand deviate from the natural rate of output. Shocks to the natural rate of output
play only a minor role in explaining the variation observed. The estimated model
suggests that monetary policy main role is to mitigate the effects of demand shocks
on aggregate demand, and to lesser extent accommodate the effect of supply shocks.
The error variance decomposition of the interest rate suggests that this is also the
case.
The impulse response functions are shown in the appendix. None of these responses
deviates from what we understand as conventional thinking, although the responses
to some of the shocks seem to be rather fast (preference shocks in particular). The
impulses from the monetary policy shock correspond well with results generated from
VARs: For a positive shock to the interest rate, the output gap falls on impact and infla-
tion reacts with a lag. The short-term interest rate falls relatively quickly and enters a
period in which the policymaker corrects for the shock.
A shock to the medium-run inflation target raises inflation expectations and the
current inflation rate on impact due to the expectations channel. The nominal interest
rate increases, but the real interest rate falls and creates a temporary increase in the
output gap which again increases inflation. Inflation peaks after five quarters and is
then brought slowly back to the steady-state rate of inflation over a 5–7 years period.
Hence, the medium-term is relatively long, approximately equal to the average busi-
ness cycle. This gives some indication of the medium-term inflation target being used

123
Natural rates in a simple New Keynesian framework 767

as an instrument to smooth output as a result of pursuing a constant inflation target

over the business cycle.
A preference shock that raises aggregate demand increases the natural real interest
rate as a higher interest rate is needed to keep output at the natural rate. The higher
natural interest rate together with increased output and inflation gaps, raise the nominal
interest rate. After an initial increase in output and inflation, both gaps fall below the
long-run equilibrium levels after four to five quarters due to the contractionary mon-
etary policy response. A cost-push shock has no influence on the natural real interest
rate, but raises inflation and lowers output in an ordinary fashion. A shock to the growth
rate of the natural rate of output raises the natural real interest rate. As people expect
income to increase permanently in the future, aggregate demand increases more than
the natural rate of output, and hence, the natural real interest rate increases. Monetary
policy reacts in a contractionary way, and the output gap is negative after having been
positive on the time of impact of the shock. Inflation is consistently below the long-run
equilibrium after the shock.

3.5 The estimated variables

The two-sided Kalman-filtered (denoted as “smoothed” in the remainder of the article)

output gap, the medium-run inflation target, and the nominal and real natural interest
rates are shown with 95% uncertainty intervals in Fig. 1. Furthermore, Fig. 2 shows
the smoothed natural rate of output and the natural real interest rate plotted with actual
output and the real interest rates, respectively, as well as the real interest rate gap
(r − r n ) and the estimated inflation gap (π̄t − π T ).
The output gap estimates suggest two recessions over the sample period: the first
one with a trough in 1991 and the other with a trough in 2001/2002. The recessions
are of approximately the same order of magnitude, suggesting a deviation of output
from the natural rate of output of approximately 5%. The recessions correspond to
periods with large positive interest-rate gaps (see Fig. 2). Further, as will be discussed
in more detail below, the dates for the turning points and the length of the business
cycles do not seem inconsistent with NBER/CBO estimates.
The dynamics of the output gap is affected by the forward-looking Phillips curve.
Inflationary pressures can be seen not only as a result of the current output gap (and
the cost-push shock), but also as a result of expected future output gaps and cost-push
shocks. This will change the dynamics relative to other measures of the output gap,
as explained under the comparisons with other estimates of the output gap.
The sample average CPI inflation over the period is 3.3%. The estimated medium-
run inflation target suggests that the mild run-up of inflation in the late 1980s, due to
a positive output gap, was partly accommodated by an increase in the inflation target
over the period, see Fig. 1. The reduction in the rate of inflation of the first part of the
1990s, accompanied by the recession in the same period, can partly be explained by a
reduction in the inflation target. From 1994 to the end of the sample, the medium-run
inflation target is estimated to be around 2% with an uncertainty band of about ±1 p.p.
For most of the period, the inflation target is significantly above zero. The inflation
gap (see Fig. 2) suggests that for the major part of the 1990s and the period after 2002,

123
768 H. C. Bjørnland et al.

Fig. 1 Inflation target, output gap, and natural interest rates. The figures show the estimated two-sided
Kalman filtered (smoothed) variables over the sample period

inflation has in general been above the medium-term inflation target, and, therefore,
has exerted an upward pressure on interest rates.
The estimate of the natural real interest rate shows considerable variation over the
period—varying between −3 and 6%. The variation in the natural real interest rate
is in periods greater than the equivalent real interest rate. This is also found in the
DSGE study of Edge et al. (2007), but not by Laubach and Williams (2003) where the
natural interest rates appear as smoothed interest rates.8 Here, the natural rate follows
instead from the stochastic processes governing the preference shocks and shocks to
the natural rate of output (see Eqs. 14 and 15). As noted above, the high degree of
prevalence of preference shocks contribute importantly to the volatility of the natural
interest rate.9 These processes are unaffected by the potential smoothing of interest

8 Using a similar model to Laubach and Williams (2003), Garnier and Wilhelmsen (2005) also find the
volatility of the natural rate of interest having decreased over time.
9 The zero-bound on nominal interest rates has been disregarded in the estimation of the model. The esti-
mate of the natural nominal interest rate becomes negative (but not significantly so) during short periods of
2002–2004. We suspect, however, that a method taking account of this constraint would not produce any
significant changes since the time periods and size of the negative interest rate are rather small.

123
 16.3 0.08

16.2
0.06

16.1
0.04

16

0.02

log GDP
15.9

Percentage points
0
15.8

15.7 −0.02

15.6 −0.04
Jan−1987 Jan−1989 Jan−1991 Jan−1993 Jan−1995 Jan−1997 Jan−1999 Jan−2001 Jan−2003 Jan−2005 Jan−1987 Jan−1989 Jan−1991 Jan−1993 Jan−1995 Jan−1997 Jan−1999 Jan−2001 Jan−2003 Jan−2005

Natural rate of output Observed output Natural real interest rate Real interest rate

0.04 0.06
Natural rates in a simple New Keynesian framework

0.04

0.02

0.02

0
0

−0.02

−0.02 −0.04
Apr−1987 Apr−1989 Apr−1991 Apr−1993 Apr−1995 Apr−1997 Apr−1999 Apr−2001 Apr−2003 Apr−2005 Jan−1987 Jan−1989 Jan−1991 Jan−1993 Jan−1995 Jan−1997 Jan−1999 Jan−2001 Jan−2003 Jan−2005

Inflation gap Interest rate gap

Fig. 2 The natural rate of output, interest-rate, and inflation gaps. From the top-left panel and moving clockwise: the natural rate of output and the real rate with actual output
and the observed real interest rate, respectively, the inflation gap (inflation minus the inflation target) and the real interest rate gap (real interest rate minus the natural real
interest rate)

123
769
770 H. C. Bjørnland et al.

3 3
2 2
1 1

Percent
Percent

0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
Jan−1986 Jan−1991 Jan−1996 Jan−2001 Jan−1986 Jan−1991 Jan−1996 Jan−2001

BLM LW BLM HP1600

3 4
2 3
1 2
1

Percent
Percent

0
0
−1
−1
−2 −2
−3 −3
−4 −4
−5 −5
Jan−1986 Jan−1991 Jan−1996 Jan−2001 Jan−1986 Jan−1991 Jan−1996 Jan−2001

BLM BN (2−sided) BLM CBO

Fig. 3 Alternative estimates of the output gap. BLM our measure of the output gap, LW an updated version
of Laubach and Williams (2003), BN the two-sided output gap estimate of Basistha and Nelson (2007),
CBO the Congressional Budget Office estimate, HP the Hodrick Prescott’s filtered output gap. See the main
text for more details

rates done by the central bank in the sticky-price equilibrium.10 Moreover, the mode
of the natural rate of interest is in the range 3–4% which does not seem unreasonable
for the average real interest rate. The average natural interest rate is remarkably stable
over the period 1994–2000 where the variation is in the region ±1 p.p. This is a result
also found by Edge et al. (2007). The recession of the first half of 2000s imply negative
real interest rates for this period, suggesting a rather expansionary monetary policy
that would have been needed to keep aggregate demand equal to the natural rate of
output.
It has been relatively common to estimate monetary policy reaction functions condi-
tional on the natural rate of interest being equal to a constant plus the inflation rate. The
relatively large variation in the natural interest rate suggests that the estimates could
be severely biased if the central bank is not taking account of the time-varying nature
of the natural rate of interest when setting interest rate. In particular, the high degree of
persistence in the natural rate in then likely to bias the coefficient on the past interest
rate upwards. Moreover, failing to take account of the interdependence between the
output gap and natural interest rates (estimates) may also bias the estimates.

10 By the same logic, there is nothing that ensures that the evolvement of the natural rate of output is
smoother than output itself. Woodford (2001, p. 234) notes “In theory, a wide variety of real shocks should
affect the growth rate of potential output[…] [T]here is no reason to assume that all of these factors follow
smooth trends. As a result, the output-gap measure that is relevant for welfare may be quite different from
simple detrended output.”

123
Natural rates in a simple New Keynesian framework 771

Table 4 Correlation and standard deviations

Estimate BLM LW HP1600 BN (2-sided) CBO

Crosscorrelations and standard deviations

BLM 1.74 0.55 0.58 0.27 0.51
LW 1.17 0.66 0.45 0.69
HP1600 0.98 0.69 0.87
BN (2-sided) 1.69 0.47
CBO 1.60
Autocorrelations
0.76 0.97 0.89 0.95 0.95

The standard deviations are shown on the diagonal of the matrix, while correlations are shown off diagonal.
See note on Fig. 3 for explanations on the gaps

Table 5 Concordance

Estimate BLM LW HP1600 BN (2-sided) CBO

BLM 1.00 0.68 0.76 0.69 0.62

LW 1.00 0.76 0.79 0.74
HP1600 1.00 0.79 0.85
BN (2-sided) 1.00 0.79
CBO 1.00

Concordance the proportion time the gaps move in the same direction. See note on Fig. 3 for explanations
on the gaps

With high volatility in the natural rate of interest, a “neutral” monetary policy
stance requires considerable changes in the interest rate. If the policymaker never-
theless regards the natural rate of interest as a constant, policy is likely to induce
inefficient movements in inflation and output.
Some readers may object to the arguments by claiming that interest rates should
be smoothed over time and, for this reason, the variability in the natural rate should
largely be ignored. We claim that such an argument mixes up two things. Interest rate
smoothing can be welfare-enhancing (see Woodford 1999) in its own right due to its
impact on private sector expectations. But optimal smoothing of interest rate does
not imply the removal of some arguments over which the smoothing should be done.
While the interest rate may be more volatile if responding to the natural interest rate,
the benefits of interest rate smoothing can still be extracted.

3.6 Alternative output gap series

We now return to the output gap in more detail, to compare our measure to some
alternative measures of the gap previously found in the literature. Figure 3 compares
our measure of the output gap (BLM henceforth) to (i) the output gap derived from an

123
772 H. C. Bjørnland et al.

updated version (2006) of Laubach and Williams (2003) (LW henceforth),11 (ii) the
two-sided output gap estimate of Basistha and Nelson (2007) (BN henceforth),12 (iii)
the Congressional Budget Office (CBO) estimate of potential output as well as (iv)
the Hodrick Prescott’s filtered output gap, with the smoothing parameter set to 1600
(HP henceforth).13 Tables 4 and 5 finally show, respectively, the correlation and the
concordance (i.e., the time proportion that the cycles of two series spend in the same
phase, see McDermott and Scott 2000)14 between the different estimates.
Our output gap series is picking the major NBER recession periods (of 1991 and
2001) efficiently. The gap is also broadly consistent with that of the other gaps, although
there are notable differences. Differences are hardly surprising given that our estimate
is consistent with a rational expectation’s forward-looking Phillips curve, whereas the
others are not. Our estimate has the highest volatility and the smallest persistence of
the series. Our Phillips curve allows for longer lags, and this implies that, for a given
value of κ, the output gap needs to move more to have the same effect on inflation. The
high degree of inflation persistence in the Phillips curve also implies that the needed
persistence in the output gap is lower to explain the observed persistence in the infla-
tion. Inflation is also more responsive to persistent changes in the output gap due to
the large coefficient on future expected inflation in the Phillips curve. In order for
the model to match inflation dynamics and volatility, the output gap then needs to be
somewhat less persistent compared to a situation with a smaller forward-looking term
in the Phillips curve. The deviations from the other series are likely to be attributable
to the differences needed for the output gap to better reflect underlying marginal costs,
as discussed above.
The differences show up in the measures of correlation. Table 4 indicates that there
is modest degree of co-movement, with correlation coefficients varying around 0.5.
The lowest correlation is found between our estimate (BLM) and that of BN.15 This
is partly explained by the early 1990s, when all the output gaps except the BN output
gap increase, with our measure suggesting a pronounced peak in 1994. Our estimate
of the natural rate of interest rose sharply over the period 1993–1995 and the interest
rate gap became negative (ref. Table 2). An expansionary monetary policy contrib-
uted to the output gap peak. The measures of concordance in the output gap, stated in
Table 5, are slightly larger than the correlation coefficient for the alternative estimates.
This implies that the estimates differ more in their sizes than their phases, that is, the
different methods tend to pick the same phase for their respective output gap estimate.
This is important information for Central Banks when comparing different gaps.

11 We thank John Williams for providing us with the updated simulation results.
12 Their output gap series was downloaded from http://www.be.wvu.edu/divecon/econ/basistha/gap.htm.
13 The Hodrick Prescott method is a univariate statistical method designed to extract the low frequency
component of a time series. Lambda penalizes the variation in the trend, and is determined a priori. A
smoothing parameter of 1600 is commonly used in many international studies.
14 The measure of concordance is useful when the focus of the analysis is on the sign of the gap and not
necessarily its magnitude.
15 In fact, the BN gap displays low correlation with all the other gaps as well.

123
Natural rates in a simple New Keynesian framework 773

4 Concluding remarks

This article provides estimates of the natural real interest rate, the output gap and the
implicit inflation target for the US economy. The inflation target since 1994 has been
remarkably stable around 2%. The natural real interest rate has, however, been varying
a lot. The assumption often made in the monetary policy literature that the natural real
interest rate is exogenous or even constant, might be very misleading and biasing the
results. For the conduct of monetary policy, acknowledging the variation in the real
interest rate and conducting policy in accordance with it, seems to be important.
By estimating the hybrid New Keynesian Phillips curve with a model-consistent
estimate of the output gap, we find that the structure of the curve is very similar to that
found by estimating the Phillips curve with the labor share of income. Our results are,
therefore, a contribution to the debate of whether it is the output gap or the labor share
of income, which provides the best representation for the inflation driving process.
If the output gap is a good representation of the inflation driving process, our results
support the idea that a simple two-variable system in inflation and the output gap (see,
Clarida et al. 1999; Woodford 2003) is a good representation of the monetary policy
transmission mechanism.

Appendix

Extra figures: Impulse response functions

See Figs. 4, 5, 6, 7, and 8.

123
774 H. C. Bjørnland et al.

Fig. 4 Monetary policy shock to the medium-term inflation target, t . The impulse response function due
to a shock to the medium-term inflation target

Fig. 5 Monetary policy shock to short-term interest rate, u t . The impulse response functions due to a shock
to the short-term interest rate

123
Natural rates in a simple New Keynesian framework 775

Fig. 6 Preference shock, ṽt . The impulse response functions due to a preference shock

Fig. 7 Cost-push shock, εt . The impulse response functions due to a cost-push shock

123
776 H. C. Bjørnland et al.

Natural real int rate −3 Output gap −3 Inflation rate

x 10 x 10
0.015 3 0

2 −0.2
0.01
1 −0.4

0 −0.6
0.005
−1 −0.8

0 −2 −1
10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60

Output growth Natural output growth −4 Nominal interest rate

x 10
0.015 0.01 20

0.008 15
0.01
0.006 10

0.004 5
0.005
0.002 0

0 0 −5
10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60

−3 Natural nom interest rate

x 10
10

−5
10 20 30 40 50 60

Fig. 8 Natural rate shock, ωt . The impulse response functions due to the natural rate of output

Acknowledgements We are grateful to Ida Wolden Bache, Leif Brubakk, Santiago Acosta Ormaechea,
Scott Schuh, and seminar participants at the 2006 Dynare Conference in Paris, the ESEM 2007 conference
in Budapest, the CEF 2007 Conference in Montréal, and the 11th ICMAIF conference in Crete 2007, and
an anonymous referee for valuable comments. We also thank John Williams for providing information from
the updated estimation of the model in Laubach and Williams (2003). The authors thank the Norwegian
Financial Market Fund under the Norwegian Research Council for financial support. Views expressed are
those of the authors and do not necessarily reflect the views of Norges Bank.

References

Adolfson M, Laséen S, Lindé J, Villani M (2007) Bayesian estimation of an open economy DSGE model
with incomplete pass-through. J Int Econ 72(2):481–511
An S, Schorfheide F (2006) Bayesian analysis of DSGE models. Working paper no 06-5, Federal Reserve
Bank of Philadelphia
Andrés J, López-Salido JD, Nelson E (2005) Sticky-price models and the natural rate hypothesis. J Monet
Econ 52:1025–1053
Basistha A, Nelson CR (2007) New measures of the output gap based on the forward-looking New Keyne-
sian Phillips curve. J Monet Econ 54:498–511
Calvo GA (1983) Staggered prices in a utility-maximizing framework. J Monet Econ 12(3):383–398
Christiano LJ, Eichenbaum M, Evans C (2005) Nominal rigidities and the dynamic effects of a shock to
monetary policy. J Polit Econ 113(1):1–45
Clarida R, Galí J, Gertler M (1999) The science of monetary policy: a New Keynesian perspective. J Econ
Lit 37(4):1661–1707
Edge RM, Kiley MT, Laforte J-P (2007) Natural rate measures in an estimated DSGE model of the U.S.
economy. Finance and Economics Discussion Series 2007–2008, Divisions of Research & Statistics
and Monetary Affairs, Federal Reserve Board

123
Natural rates in a simple New Keynesian framework 777

Fuhrer JC (2000) Habit formation in consumption and its implications for monetary-policy models. Am
Econ Rev 90(3):367–390
Galí J, Gertler M (1999) Inflation dynamics: a structural econometric analysis. J Monet Econ 44:195–222
Galí J, Gertler M, López-Salido JD (2001) European inflation dynamics. Eur Econ Rev 45(7):1237–1270
Galí J, Gertler M, López-Salido D (2003) Erratum to ‘European inflation dynamics’. Eur Econ Rev
47(4):759–760
Galí J, Gertler M, Lopez-Salido DJ (2005) Robustness of the estimates of the hybrid New Keynesian Phillips
curve. J Monet Econ 52(6):1107–1118
Garnier J, Wilhelmsen B-R (2005) The natural real interest rate and the output gap in the Euro area. Europena
Central Bank working paper no. 546
Goodfriend M, King RG (1997) The new neoclassical synthesis and the role of monetary policy. In:
Bernanke BS, Rotemberg JJ (eds) NBER macroeconomics annual. MIT Press, Cambridge,
pp 231–283
Gordon RJ (1998) Foundations of the goldilocks economy: supply shocks and the time-varying NAIRU.
Brookings Pap Econ Act 0(2):297–333
Harvey AC (1989) Forecasting, structural time series models and the Kalman filter. Cambridge University
Press, New York
Ireland PN (2007) Changes in the federal reserve’s inflation target: causes and consequences. Manuscript,
Boston College
Juillard M, Karam P, Laxton D, Pesenti P (2005) Welfare-based monetary policy rules in an estimated
DSGE model of the US economy. Manuscript, Federal Reserve Bank of New York
Juillard M, Kamenik O, Kumhof M, Laxton D (2006) Measures of potential output from an estimated DSGE
model of the United States. Manuscript, CEPREMAP and IMF
Laubach T, Williams JC (2003) Measuring the natural rate of interest. Rev Econ Stat 85(4):1063–1070
McCallum BT, Nelson E (1999) An optimizing IS-LM specification for monetary policy and business cycle
analysis. J Money Credit Bank 31(3):296–316. NBER working paper no. W5875
McDermott JC, Scott A (2000) Concordance in business cycles. IMF working papers 00/37
Rotemberg JJ, Woodford M (1997) An optimizing-based econometric model for the evaluation of monetary
policy. In: Rotemberg JJ, Bernanke BS (eds) NBER macroeconomics annual. MIT Press, Cambridge,
pp 297–346
Rudebusch G (2002a) Term structure evidence on interest rate smoothing and monetary policy inertia. J
Monet Econ 49:1161–1187
Rudebusch GD (2002b) Assessing nominal income rules for monetary policy with model and data uncer-
tainty. Econ J 112:1–31
Sbordone AM (2002) Prices and unit labor costs: a new test of price stickiness. J Monet Econ 49(2):265–292
Sbordone AM (2005) Do expected future marginal costs drive inflation dynamics? J Monet Econ 52:
1183–1197
Schorfheide F (2000) Loss function-based evaluation of DSGE models. J Appl Econom 15(6):645–670
Smets F, Wouters R (2003) An estimated dynamic stochastic general equilibrium model of the Euro area.
J Eur Econ Assoc 1:1123–1175
Smets F, Wouters R (2007) Shocks and frictions in US business cycles. Europena Central Bank working
paper no. 722
Svensson LE (1997) Inflation forecast targeting: implementing and monitoring inflation targets. Eur Econ
Rev 41:1111–1146
Taylor JB (1993) Discretion versus policy rules in practice. Carnegie Rochester Conf Ser Public Policy
39:195–214
Watson MW, Staiger D, Stock J (1997) How precise are estimates of the natural rate of unemployment. In:
Romer CD, Romer DH (eds) Reducing inflation: motivation and strategy, chap 5. The University of
Chicago Press, Chicago, pp 195–246
Woodford M (1999) Optimal monetary policy inertia. Working paper, Princeton University
Woodford M (2001) The Taylor rule and optimal monetary policy. Am Econ Rev 91(2):232–237
Woodford M (2003) Interest and prices: foundations of a theory of monetary policy. Princeton University
Press, Princeton

123